I suggest you think more about the problem... Whenever I hear someone talk about unnecessary RPC calls in a distributed system.... Well, I get skeptical. Especially in light of 10GBe.
Sloppy code is one thing. Being myopic is another. This is why I am asking for a more concrete use case. Lars makes a point that there is a wide variety of potential use cases. However trying to solve a bad use case with something that could have a negative impact on overall performance for other use cases... well, I'd like to avoid that if it were possible. I think I can give you a couple of examples... First if you had a use case where you really needed a distributed OLTP database, I'd say HBase wasn't the right tool and you should look at Informix's XPS engine,provided IBM still sells it... The point being Second, HBase sits on top of HDFS. Because of design issues in HDFS, like not having R/W access to files, there are limitations to HBase where we have to deal with things like compactions. I point this out because its a design constraint that impacts the solutions we design using HBase. In these two examples, we would want to question the use of HBase as part of the solution. This is why I'm asking for a use case of where indexing at the region level makes sense. It sounds like the idea is to use the secondary index as a filter on a range scan. Using an inverted table, the columns are in sort order. So that it would be easier and lighter to fetch only the columns within the range of the query. Simple extension to using a secondary inverted table for your index. Just saying... Sent from a remote device. Please excuse any typos... Mike Segel On Jan 8, 2013, at 10:11 PM, ramkrishna vasudevan <[email protected]> wrote: > As far as i can see its more related to using the coprocessor framework in > this soln that helps us in a great way to avoid unnecessary RPC calls when > we go with Region level indexing. > > Regards > Ram > > On Wed, Jan 9, 2013 at 8:52 AM, Anoop Sam John <[email protected]> wrote: > >> Totally agree with Lars. The design came up as per our usage and data >> distribution style etc. >> Also the put performance we were not able to compromise. That is why the >> region collocation based region based indexing design came :) >> Also as we are having the indexing and index usage every thing happening >> at server side, there is no need for any change in the client part >> depending on what type of client u use. Java code or REST APIs or any >> thing. Also MR based parallel scans any thing can be comparably easy I >> feel as there is absolutely no changes needed at client side. :) >> >> As Anil said there will be pros and cons for every way and which one suits >> your usage, needs to be adopted. :) >> >> -Anoop- >> ________________________________________ >> From: anil gupta [[email protected]] >> Sent: Wednesday, January 09, 2013 6:58 AM >> To: [email protected]; lars hofhansl >> Subject: Re: HBase - Secondary Index >> >> +1 on Lars comment. >> >> Either the client gets the rowkey from secondary table and then gets the >> real data from Primary Table. ** OR ** Send the request to all the RS(or >> region) hosting a region of primary table. >> >> Anoop is using the latter mechanism. Both the mechanism have their pros and >> cons. IMO, there is no outright winner. >> >> ~Anil Gupta >> >> On Tue, Jan 8, 2013 at 4:30 PM, lars hofhansl <[email protected]> wrote: >> >>> Different use cases. >>> >>> >>> For global point queries you want exactly what you said below. >>> For range scans across many rows you want Anoop's design. As usually it >>> depends. >>> >>> >>> The tradeoff is bringing a lot of unnecessary data to the client vs >> having >>> to contact each region (or at least each region server). >>> >>> >>> -- Lars >>> >>> >>> >>> ________________________________ >>> From: Michael Segel <[email protected]> >>> To: [email protected] >>> Sent: Tuesday, January 8, 2013 6:33 AM >>> Subject: Re: HBase - Secondary Index >>> >>> So if you're using an inverted table / index why on earth are you doing >> it >>> at the region level? >>> >>> I've tried to explain this to others over 6 months ago and its not really >>> a good idea. >>> >>> You're over complicating this and you will end up creating performance >>> bottlenecks when your secondary index is completely orthogonal to your >> row >>> key. >>> >>> To give you an example... >>> >>> Suppose you're CCCIS and you have a large database of auto insurance >>> claims that you've acquired over the years from your Pathways product. >>> >>> Your primary key would be a combination of the Insurance Company's ID and >>> their internal claim ID for the individual claim. >>> Your row would be all of the data associated to that claim. >>> >>> So now lets say you want to find the average cost to repair a front end >>> collision of an S80 Volvo. >>> The make and model of the car would be orthogonal to the initial key. >> This >>> means that the result set containing insurance records for Front End >>> collisions of S80 Volvos would be most likely evenly distributed across >> the >>> cluster's regions. >>> >>> If you used a series of inverted tables, you would be able to use a >> series >>> of get()s to get the result set from each index and then find their >>> intersections. (Note that you could also put them in sort order so that >> the >>> intersections would be fairly straight forward to find. >>> >>> Doing this at the region level isn't so simple. >>> >>> So I have to again ask why go through and over complicate things? >>> >>> Just saying... >>> >>> On Jan 7, 2013, at 7:49 AM, Anoop Sam John <[email protected]> wrote: >>> >>>> Hi, >>>> It is inverted index based on column(s) value(s) >>>> It will be region wise indexing. Can work when some one knows the >> rowkey >>> range or NOT. >>>> >>>> -Anoop- >>>> ________________________________________ >>>> From: Mohit Anchlia [[email protected]] >>>> Sent: Monday, January 07, 2013 9:47 AM >>>> To: [email protected] >>>> Subject: Re: HBase - Secondary Index >>>> >>>> Hi Anoop, >>>> >>>> Am I correct in understanding that this indexing mechanism is only >>>> applicable when you know the row key? It's not an inverted index truly >>>> based on the column value. >>>> >>>> Mohit >>>> On Sun, Jan 6, 2013 at 7:48 PM, Anoop Sam John <[email protected]> >>> wrote: >>>> >>>>> Hi Adrien >>>>> We are making the consistency btw the main table and >>>>> index table and the roll back mentioned below etc using the CP hooks. >>> The >>>>> current hooks were not enough for those though.. I am in the process >> of >>>>> trying to contribute those new hooks, core changes etc now... Once >> all >>> are >>>>> done I will be able to explain in details.. >>>>> >>>>> -Anoop- >>>>> ________________________________________ >>>>> From: Adrien Mogenet [[email protected]] >>>>> Sent: Monday, January 07, 2013 2:00 AM >>>>> To: [email protected] >>>>> Subject: Re: HBase - Secondary Index >>>>> >>>>> Nice topic, perhaps one of the most important for 2013 :-) >>>>> I still don't get how you're ensuring consistency between index table >>> and >>>>> main table, without an external component (such as >>> bookkeeper/zookeeper). >>>>> What's the exact write path in your situation when inserting data ? >>>>> (WAL/RegionObserver, pre/post put/WALedit...) >>>>> >>>>> The underlying question is about how you're ensuring that WALEdit in >>> Index >>>>> and Main tables are perfectly sync'ed, and how you 're able to >> rollback >>> in >>>>> case of issue in both WAL ? >>>>> >>>>> >>>>> On Fri, Dec 28, 2012 at 11:55 AM, Shengjie Min <[email protected]> >>>>> wrote: >>>>> >>>>>>> Yes as you say when the no of rows to be returned is becoming more >> and >>>>>> more the latency will be becoming more. seeks within an HFile block >> is >>>>>> some what expensive op now. (Not much but still) The new encoding >>>>>> prefix >>>>>> trie will be a huge bonus here. There the seeks will be flying.. [Ted >>>>> also >>>>>> presented this in the Hadoop China] Thanks to Matt... :) I am >> trying >>> to >>>>>> measure the scan performance with this new encoding . Trying to >back >>>>> port >>>>>> a simple patch for 94 version just for testing... Yes when the no >> of >>>>>> results to be returned is more and more any index will become less >>>>>> performing as per my study :) >>>>>> >>>>>> yes, you are right, I guess it's just a drawback of any index >> approach. >>>>>> Thanks for the explanation. >>>>>> >>>>>> Shengjie >>>>>> >>>>>> On 28 December 2012 04:14, Anoop Sam John <[email protected]> >> wrote: >>>>>> >>>>>>>> Do you have link to that presentation? >>>>>>> >>>>>>> http://hbtc2012.hadooper.cn/subject/track4TedYu4.pdf >>>>>>> >>>>>>> -Anoop- >>>>>>> >>>>>>> ________________________________________ >>>>>>> From: Mohit Anchlia [[email protected]] >>>>>>> Sent: Friday, December 28, 2012 9:12 AM >>>>>>> To: [email protected] >>>>>>> Subject: Re: HBase - Secondary Index >>>>>>> >>>>>>> On Thu, Dec 27, 2012 at 7:33 PM, Anoop Sam John <[email protected] >>> >>>>>>> wrote: >>>>>>> >>>>>>>> Yes as you say when the no of rows to be returned is becoming more >>>>> and >>>>>>>> more the latency will be becoming more. seeks within an HFile >> block >>>>> is >>>>>>>> some what expensive op now. (Not much but still) The new encoding >>>>>> prefix >>>>>>>> trie will be a huge bonus here. There the seeks will be flying.. >> [Ted >>>>>>> also >>>>>>>> presented this in the Hadoop China] Thanks to Matt... :) I am >>>>> trying >>>>>> to >>>>>>>> measure the scan performance with this new encoding . Trying to >> back >>>>>>> port a >>>>>>>> simple patch for 94 version just for testing... Yes when the no >> of >>>>>>>> results to be returned is more and more any index will become less >>>>>>>> performing as per my study :) >>>>>>>> >>>>>>>> Do you have link to that presentation? >>>>>>> >>>>>>> >>>>>>>>> btw, quick question- in your presentation, the scale there is >>>>> seconds >>>>>> or >>>>>>>> mill-seconds:) >>>>>>>> >>>>>>>> It is seconds. Dont consider the exact values. What is the % of >>>>>> increase >>>>>>>> in latency is important :) Those were not high end machines. >>>>>>>> >>>>>>>> -Anoop- >>>>>>>> ________________________________________ >>>>>>>> From: Shengjie Min [[email protected]] >>>>>>>> Sent: Thursday, December 27, 2012 9:59 PM >>>>>>>> To: [email protected] >>>>>>>> Subject: Re: HBase - Secondary Index >>>>>>>> >>>>>>>>> Didnt follow u completely here. There wont be any get() >> happening.. >>>>>> As >>>>>>>> the >>>>>>>>> exact rowkey in a region we get from the index table, we can seek >> to >>>>>> the >>>>>>>>> exact position and return that row. >>>>>>>> >>>>>>>> Sorry, When I misused "get()" here, I meant seeking. Yes, if it's >>>>> just >>>>>>>> small number of rows returned, this works perfect. As you said you >>>>> will >>>>>>> get >>>>>>>> the exact rowkey positions per region, and simply seek them. I was >>>>>> trying >>>>>>>> to work out the case that when the number of result rows increases >>>>>>>> massively. Like in Anil's case, he wants to do a scan query against >>>>> the >>>>>>>> 2ndary index(timestamp): "select all rows from timestamp1 to >>>>>> timestamp2" >>>>>>>> given no customerId provided. During that time period, he might >> have >>>>> a >>>>>>> big >>>>>>>> chunk of rows from different customerIds. The index table returns a >>>>> lot >>>>>>> of >>>>>>>> rowkey positions for different customerIds (I believe they are >>>>>> scattered >>>>>>> in >>>>>>>> different regions), then you end up seeking all different positions >>>>> in >>>>>>>> different regions and return all the rows needed. According to your >>>>>>>> presentation page14 - Performance Test Results (Scan), without >> index, >>>>>>> it's >>>>>>>> a linear increase as result rows # increases. on the other hand, >> with >>>>>>>> index, time spent climbs up way quicker than the case without >> index. >>>>>>>> >>>>>>>> btw, quick question- in your presentation, the scale there is >> seconds >>>>>> or >>>>>>>> mill-seconds:) >>>>>>>> >>>>>>>> - Shengjie >>>>>>>> >>>>>>>> >>>>>>>> On 27 December 2012 15:54, Anoop John <[email protected]> >> wrote: >>>>>>>> >>>>>>>>>> how the massive number of get() is going to >>>>>>>>> perform againt the main table >>>>>>>>> >>>>>>>>> Didnt follow u completely here. There wont be any get() >> happening.. >>>>>> As >>>>>>>> the >>>>>>>>> exact rowkey in a region we get from the index table, we can seek >>>>> to >>>>>>> the >>>>>>>>> exact position and return that row. >>>>>>>>> >>>>>>>>> -Anoop- >>>>>>>>> >>>>>>>>> On Thu, Dec 27, 2012 at 6:37 PM, Shengjie Min < >>>>> [email protected]> >>>>>>>>> wrote: >>>>>>>>> >>>>>>>>>> how the massive number of get() is going to >>>>>>>>>> perform againt the main table >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> All the best, >>>>>>>> Shengjie Min >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> All the best, >>>>>> Shengjie Min >>>>> >>>>> >>>>> >>>>> -- >>>>> Adrien Mogenet >>>>> 06.59.16.64.22 >>>>> http://www.mogenet.me >> >> >> >> >> -- >> Thanks & Regards, >> Anil Gupta >>
